📄 中文摘要
运动规划在狭窄通道中的应用仍然是一个核心挑战。基于采样的规划器很少将样本放置在这些狭窄但关键的区域内，即使样本落在通道内，样本之间的直线连接也常常靠近障碍物边界，因而频繁被碰撞检测拒绝。分解式规划器通过将自由空间划分为凸单元来解决这两个问题，确保每个通道都被精确捕捉为单元边界，且单元内的任何路径在构造上都是无碰撞的。然而，随着环境复杂性的增加，候选走廊的数量呈组合增长，导致走廊选择成为瓶颈。GNN-DIP框架提出了一种解决方案，旨在提高走廊选择的效率。
GNN-DIP：基于神经网络的分解式运动规划中的走廊选择
📄 English Summary
GNN-DIP: Neural Corridor Selection for Decomposition-Based Motion Planning
Motion planning through narrow passages remains a significant challenge. Sampling-based planners rarely place samples in these critical regions, and even when samples do land within a passage, the straight-line connections between them often run close to obstacle boundaries, leading to frequent rejection by collision checking. Decomposition-based planners address these issues by partitioning free space into convex cells, ensuring that every passage is captured as a cell boundary and any path within a cell is collision-free by construction. However, the number of candidate corridors through the cell graph grows combinatorially with the complexity of the environment, creating a bottleneck in corridor selection. The GNN-DIP framework presents a solution aimed at improving the efficiency of corridor selection.